Dietary underreporting by obese individuals--is it specific or non-specific?BMJ 1995; 311 doi: https://doi.org/10.1136/bmj.311.7011.986 (Published 14 October 1995) Cite this as: BMJ 1995;311:986
- a Danish Epidemiology Science Centre, Institute of Preventive Medicine, Copenhagen, and Glostrup Population Studies, Medical Department C, Glostrup Hospital, Glostrup, Denmark
- b Department of Internal Medicine and Primary Health Care, Gothenburg University, Gothenburg, Sweden
- Correspondence to: Dr B L Heitmann, Institute of Preventive Medicine, Copenhagen Hospital Corporation, Copenhagen Municipal Hospital, DK-1399 Copenhagen K, Denmark.
Objective: To examine the distribution of patterns of macronutrient density in relation to obesity.
Design: Cross sectional.
Subjects: 323 men and women aged 35-65 years, selected randomly from a larger population sample of Danish adults.
Main outcome measure: Bias in dietary reporting of energy and protein intake in relation to percentage body fat, assessed by comparison of data from an interview on dietary intake with data estimated from 24 hour nitrogen output, validated by administering p-aminobenzoic acid, and estimated 24 hour energy expenditure.
Results: Degree of obesity was positively associated with underreporting of total energy and protein, whereas compared with total energy reported, protein was overreported by the obese subjects.
Conclusion: Errors in dietary reporting of protein seem to occur disproportionately with respect to total energy, suggesting a differential reporting pattern of different foods. Although, on average, all subjects showed a greater underreporting of energy than of protein, this was most common in the obese subjects. Snack-type foods may be preferentially forgotten when obese people omit food items in dietary reporting. These results seem to agree with the general assumption that obese people tend to underreport fatty foods and foods rich in carbohydrates rather than underreport their total dietary intake. These results may have implications for the interpretation of studies of diet and comorbidities related to obesity.
Actual intake of energy from protein can be calculated from 24 hour urinary nitrogen output and energy expenditure
This study shows that obese subjects underreport both total energy and protein intake, with 15% more underreporting of energy than protein
The results have implications for interpreting studies of effects of dietary intake on comorbidities related to obesity
Future research is needed to identify valid biomarkers for total fat and complex and simple carbohydrates
Associations between diet and health are often weak or inconsistent. This is generally attributed to difficulties in obtaining valid information on dietary intake. Several reports have shown that obese individuals underreport their dietary intake substantially, whether the intake is self recorded or assessed by interview.1 2 3 4 5 Whether this is a consequence of selective underreporting of certain foods (such as snacks or items considered unhealthy) or general underreporting of all food, or both, is not known. Methods such as the doubly labelled water technique,5 6 or calculating 24 hour urinary nitrogen output,7 8 have been used to describe the magnitude of underreporting. Such errors in dietary reporting may not invalidate the dietary data, provided that the underreporting is not selective. Indeed, comparisons of methods of reporting dietary intake in obese subjects have shown similar patterns of nutrient density despite substantial differences in total energy intake shown by different methods.9 Such results, however, are limited by the fact that no method of reporting dietary intake can be assumed to be unbiased. We examined the errors in patterns of nutrient density in relation to obesity by comparing data from an interview on dietary intake with data on urinary nitrogen output and estimated energy expenditure.
Subjects and methods
The study was part of the Danish MONICA project (an international study conducted under the auspices of the World Health Organisation to monitor trends in and determinants of mortality from cardiovascular disease) carried out between 1987 and 1988.10 11 The study population included 552 Danish citizens aged 35, 45, 55, and 65 years, selected at random from a larger population sample of 4581 subjects.10
The project was approved by the ethics committee for Copenhagen county and is in accordance with the Helsinki II declaration on human rights.
All participants answered questions about physical activity during leisure (sitting most of the time; light activity at least four hours a week; active in sport at least three hours a week or heavy work during leisure; active in competitive sport several times a week) and physical activity during work (no work; sitting most of the time; light activity, walking around; walking and carrying most of the time; strenuous physical work).
One trained dietician interviewed all subjects by taking a diet history. Average daily intakes were calculated from responses about the previous month. Data on meal patterns, dishes, and food items were obtained with a precoded interview form. Quantities were assessed with food models, series of photographs, cups, and measures. Calculations of nutrients were carried out with the DANKOST program, which is derived from the Danish food composition tables.12
COLLECTION AND ANALYSIS OF URINE
All subjects were instructed orally and in writing on collecting 24 hour urine samples. To monitor completeness, each participant was given three tablets containing 80 mg p-aminobenzoic acid to take during the day of urine collection.13 Protein intake was calculated from the 24 hour nitrogen output according to the formula of Isaksson7:
Calculated protein intake based on urine samples (g) = (Nitrogen output in 24 hour urine (g/day) + 2 g) x 6.25
The 30 µl urine samples were analysed for nitrogen by flash combustion technique (NA 1500 Nitrogen Analyser, Ciba Corning13).
All anthropometric measurements were taken in accordance with the WHO's standards.14 Height was measured to the nearest 0.5 cm. Body weight was measured to the nearest 0.1 kg (SECA weighing scales), with the subjects wearing only underwear.
MEASUREMENT OF ELECTRICAL IMPEDANCE
A BIA 103 RJL system analyser (RJL, Detroit) was used to measure electrical impedance, according to the instructions given by the manufacturer. The measurements were taken with a tetrapolar electrode placement, with electrodes placed on the right hand and foot.
The algorithm used to estimate body fat from impedance had been developed in a subgroup of the same sample of Danes15:
Body fat (kg)=(0.819xbody weight (kg))-(0.279xheight2/resistance (cm2/ohm))-(0.064xsexxbody weight (kg))+(0.077xage (years))-(0.231xheight (cm)) + 14.941
where sex is coded as 1 for men and 0 for women.
Basal energy expenditure was calculated according to the formula of Gadoy et al16:
Basal energy expenditure (J)=(116.76xfat free mass)+(26.88xbody fat)
where fat free mass is the difference between body weight and body fat.
Average 24 hour energy expenditure can be expressed as a multiple of basal energy expenditure, known as the physical activity level. The physical activity level was assumed to be 1.55 (men) and 1.56 (women) among those who were unemployed or whose work was classified as sedentary; 1.78 and 1.64 respectively among those whose work was classified as light activity; and 2.10 and 1.82 respectively among those engaged in heavy and strenuous work or engaged actively in sports.14
Multiple regression and partial correlation were used to describe associations between reporting error for energy and protein, and degree of obesity. The F test was used to examine differences in the mean and in regression coefficients between groups, as described earlier.17 Reporting bias of energy and protein, adjusted for covariates, by percentage body fat in tenths was calculated for men and women separately, with analysis of variance.
Of the 552 subjects invited (276 men and 276 women), 476 attended the health examination, and 435 agreed to participate in the study. In all, 386 subjects gave both the 24 hour urine sample and the dietary interview. The 63 subjects whose urine samples contained less than 85% of the administered p-aminobenzoic acid were excluded from further analysis.9 Thus 323 subjects remained in the study. As described earlier1 no difference in recovery of p-aminobenzoic acid was found between age groups. Slightly more women than men had acceptable compliance of p-aminobenzoic acid (P=0.06). Within each sex, however, no differences were found in age, body weight, height, body mass index, percentage body fat, or prevalence of obesity (body mass index >30 kg/m2) among (a) the 63 subjects who were interviewed about dietary intake but gave an incomplete urine sample; (b) the 49 subjects who did not return a urine sample; and (c) the 323 subjects who both were interviewed and gave a complete sample (all, P>0.30).
Table I shows the subjects' characteristics. In both men and women, estimates of total energy and total protein intake calculated from the dietary interview were lower than those calculated from the urine analyses (all, P<0.0001), whereas estimates of percentage energy from protein calculated from the dietary interview were significantly higher than estimates based on urinary nitrogen (all, P<0.001). In all, 85% (275/323) and 72% (233/323) of the subjects underreported total energy and protein respectively.
Reported energy and protein intakes were negatively associated, estimated 24 hour energy expenditure was positively associated, and intake of protein (calculated from urinary nitrogen output) was unassociated with obesity (table II). Relative to total energy, protein was overreported by obese subjects, as suggested by the stronger association for reporting of energy than for reporting of protein. Similar associations were seen when recalculating the data after energy from alcohol intake was omitted (associations between underreported protein relative to underreported energy and percentage body fat, r=0.25, P<0.0001). Differences between men and women were not significant (all, P<0.15), although associations tended to be stronger for women.
Figure 1 shows percentage body fat in relation to total energy and protein intake (estimated from the diet), energy expenditure, and urinary nitrogen output. In both men and women the difference between reported and estimated intake of total energy and protein increased with increasing percentage body fat (all, P<0.03). With increasing obesity the proportion of underreported energy compared to underreported protein increased. Underreporting of energy in subjects in the highest tenth in terms of percentage body fat was 13% greater among men and 15% greater among women than for protein (fig 2).
This study shows that the degree of obesity influences dietary reporting both quantitatively and qualitatively. In agreement with other studies,1 2 3 4 a direct relation was seen between obesity and dietary underreporting of energy and protein. With obesity, however, underreporting was proportionally greater, and increasingly greater for energy than for protein, suggesting a differential dietary reporting pattern with respect to different foods. Thus the results seem to support the general assumption that obese people tend to underreport fatty foods or foods rich in carbohydrates, or both, rather than underreport all constituents.
Only a few biological markers of dietary intake are available, and none of these validates intakes of total fat or carbohydrates. Although this study shows that obese people tend to overreport relative energy from protein intake, it does not show to what degree, or in what proportion, reporting bias occurred for these other macronutrients. Furthermore, that obesity might influence reporting of alcohol intake cannot be excluded. The fact, however, that similar associations remained between reporting bias and obesity, without energy from alcohol included in the analyses, implies differences of energy densities from other macronutrients.
Reporting bias tended to be stronger for women than for men, although not significantly so. This finding is consistent with the observations of Drewnowsky et al18 that obese men prefer protein and fat mixtures while women prefer carbohydrate and fat mixtures. Thus the overreporting of protein is probably easier to detect in women.
The question of specific errors in dietary reporting has been addressed in other population studies,9 19 20 21 but results are inconsistent. One study suggested selective underreporting of snacks,19 while another showed no selective underreporting20 but did not consider obesity. Comparing results of different methods of assessing dietary intake among obese subjects, one study showed similar densities of macronutrients9 and another showed different densities,21 despite large differences in energy intake, indicating that method specific biases probably differ among populations.
POSSIBLE LIMITATIONS OF STUDY
This study has potential limitations. Selective overreporting of physical activity by obese people could invalidate our findings. However, recalculating associations between percentage body fatand underreported protein relative to underreported energy--assuming that all obese subjects (top third) were physically inactive (physical activity level 1.56)--gave virtually identical values (r=0.15 v r=0.20) but an upward shift of the intercept, suggesting that reporting pattern of physical activity was similar in lean and obese subjects. Unbiased calculation of 24 hour energy expenditure is crucial for estimating percentage energy from protein from urinary nitrogen output. Rather than estimating 24 hour energy expenditure from height and weight, we used sex specific and age specific equiations based on body fat, fat free mass, and physical activity during work and leisure.14 16 In studies using doubly labelled water or indirect calorimetry, body fat and fat free mass have been found to explain more than 90% of the variation in energy expenditure.22 23 At the individual level, calculation of oxygen consumption by using doubly labelled water would be expected to give a more accurate estimate of energy expenditure, by accounting for genetic variation in 24 hour energy expenditure, residual variation in voluntary physical activity level, and bias in reporting of physical activity by obese people24 not captured by the relatively crude activity measure. The results reported here should therefore be confirmed by using doubly labelled water to determine actual free living 24 hour energy expenditure.
The results of this study suggest that imprecise dietary reporting can be both a quantitative and a qualitative phenomenon. The quantitative dimension, which has been shown in obese populations,1 2 3 4 5 may be a direct function of the quantity of food consumed, making it natural that those obese people who eat the most underreport to a greater extent.25 On the other hand, the qualitative dimension is independent of the quantity of energy consumed. The fact that this bias does not seem to be present when contrasting methods of dietary reporting9 suggests that selective underreporting may occur even with methods that are relatively efficient at capturing total energy intake.26
Our results have implications for the interpretation of dietary surveys and will be relevant in studies of effects of dietary intake on comorbidities related to obesity. The results are consistent with the idea that even macronutrient densities may be biased with respect to obesity. If, in fact, the proportion of fat is differentially underreported by obese individuals, this would imply that previous studies showing positive associations between percentage energy from fat intake and health outcomes related to obesity may have overestimated the true effects. Future research is needed to identify valid dietary biomarkers for total fat and complex and simple carbohydrates. Such techniques are necessary to understand fully the nature of the bias in the Danish population sample shown here.
This work was done in collaboration with the research department of human nutrition of the Royal Veterinary and Agricultural University, Denmark. The authors thank Drs Bjorn Isaksson, Erling S Olesen, and Morten Gronbaek for inspiration and helpful comments on the manuscript.
Funding Funding: The project was supported by grants from the Danish Agricultural and Veterinary Research and Danish Medical Research Councils and the Danish Health Insurance Foundation.
Conflict of interest None